This article was downloaded by: [134.117.10.

200] On: 08 August 2020, At: 03:41

Publisher: Institute for Operations Research and the Management Sciences (INFORMS)
INFORMS is located in Maryland, USA

Management Science
Publication details, including instructions for authors and subscription information:
[Link]

Knowledge Sharing in Alliances and Alliance Portfolios

Ashish Arora, Sharon Belenzon, Andrea Patacconi

To cite this article:

Ashish Arora, Sharon Belenzon, Andrea Patacconi (2020) Knowledge Sharing in Alliances and Alliance Portfolios. Management
Science

Published online in Articles in Advance 06 Aug 2020

. [Link]

Full terms and conditions of use: [Link]

Conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial use
or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher
approval, unless otherwise noted. For more information, contact permissions@[Link].

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitness
for a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, or
inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or
support of claims made of that product, publication, or service.

Copyright © 2020, INFORMS

Please scroll down for article—it is on subsequent pages

With 12,500 members from nearly 90 countries, INFORMS is the largest international association of operations research (O.R.)
and analytics professionals and students. INFORMS provides unique networking and learning opportunities for individual
professionals, and organizations of all types and sizes, to better understand and use O.R. and analytics tools and methods to
transform strategic visions and achieve better outcomes.
For more information on INFORMS, its publications, membership, or meetings visit [Link]
 MANAGEMENT SCIENCE
Articles in Advance, pp. 1–17
[Link] ISSN 0025-1909 (print), ISSN 1526-5501 (online)

Knowledge Sharing in Alliances and Alliance Portfolios

Ashish Arora,a,b Sharon Belenzon,a,b Andrea Patacconic
a
Fuqua School of Business, Duke University, Durham, North Carolina 27708 b National Bureau of Economic Research, Cambridge,
Massachusetts 02138 c Norwich Business School, University of East Anglia, Norwich NR4 7TJ, United Kingdom
Contact: [Link]@[Link] (AA); [Link]@[Link], [Link] (SB); [Link]@[Link],
[Link] (AP)

Received: May 15, 2017 Abstract. We develop a model of knowledge sharing in alliances and alliance portfolios.
Revised: January 7, 2019; August 23, 2019 We show that, once the issue of encouraging effective collaboration is put center stage,
Accepted: January 12, 2020 many standard intuitions of the learning race view and alliance portfolio literature are
Published Online in Articles in Advance: overturned or qualified. Partners engage in learning races in some cases, but exhibit
August 6, 2020 “altruistic” behaviors in other cases. They may reduce their own absorptive capacity or
[Link] increase the transparency of their own operations to facilitate their partner’s learning. In
alliance portfolios, we show that not all substitutability between alliance portfolio partners
Copyright: © 2020 INFORMS is bad. We distinguish between substitutability in implementation and substitutability in
rival benefits and show that the latter is conducive to knowledge sharing. Our work
contributes toward putting the literature on learning alliances on a more solid founda-
tion by emphasizing the importance of commitments that leading firms can make to
encourage collaboration.

History: Accepted by David Simchi-Levi, business strategy.

Supplemental Material: The online appendix is available at [Link] 10.1287/mnsc.2020.3614.

Keywords: knowledge sharing • learning alliances • alliance portfolios • commitment • learning races

1. Introduction (Christensen 1994). The goal of this alliance was to

Learning alliances, in which an important objective of develop single-use lens molds for disposable contact
the partners is the creation of new knowledge, are lenses. Although the project required substantial joint
difficult to manage. On the one hand, the partnering problem solving, conflicts arose when attempts were
firms (which may also be competitors) must share made to more fully integrate the two firms’ engi-
knowledge to create new knowledge and cooperatively neering teams. Senior managers at Nypro opposed
exploit this jointly created new knowledge (Hennart these attempts on the grounds that they could com-
1988, Mitchell and Singh 1996). This gives them a promise Nypro’s intellectual property and bargain-
common purpose. On the other hand, the partners ing position. Christensen highlights three key con-
must divide the gains from collaboration. The nego- cerns. First, Nypro managers noted that, although
tiations are affected by the relative bargaining posi- Nypro held several patents, much of its technology
tions after sharing, which depend upon how efficiently was not patented as it involved process innovations
firms have learned from each other and how well po- that were best protected through trade secrets. One
sitioned they are to exploit that learning without the concern was that Vistakon engineers, after learning
cooperation of the partner (Hamel 1991, Yan and Gray trade secrets from Nypro, could take jobs with injec-
1994, Panico 2017). tion molders directly competing with Nypro. A second
In many cases, the new knowledge created cannot concern was that Vistakon could use this information
be accurately foreseen nor adequately distinguished to develop alternative technologies to making lenses,
from what is already known by the partners. Thus, which Vistakon was free to explore in parallel with its
firms cannot contract or commit to jointly exploit it. program with Nypro. Third, Nypro worried that, if the
Any existing contract or joint venture terms are subject project was successful, Vistakon could use its better
to renegotiation under the threat that one partner or understanding of Nypro’s manufacturing process to
the other may walk away (Williamson 1975, Grossman constrain Nypro’s pricing flexibility and bargaining
and Hart 1986). As a result, firms may be reluctant power in future supply contracts. For these reasons,
to collaborate and share knowledge ex ante if they knowledge sharing with Vistakon had to be managed
expect to find themselves in a weak bargaining po- carefully and sometimes constrained.
sition ex post. Naturally alliance partners may attempt to mitigate
Consider, for instance, the alliance between Nypro such risks from knowledge sharing. Much of the lit-
and Vistakon, a division of Johnson & Johnson erature focuses on defense mechanisms that firms can

1
 Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
2 Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS

use to protect their knowledge from expropriation are reduced. Thus, substitutability in implementation
(e.g., Cohen et al. 2000, Katila et al. 2008). Here we hinders knowledge sharing.
take a complementary perspective and study orga- Substitutability in rival benefits, by contrast, refers
nizational commitments that leading firms can use to to a situation in which the focal firm F does not gain
encourage knowledge sharing when appropriability much by appropriating the knowledge of both A
hazards are present (Schelling 1960, Williamson 1983). and B relative to appropriating the knowledge of
Nypro was concerned that sharing its knowledge, only A or B. Intuitively, if A’s and B’s knowledge bases
though vital to the success of the alliance, would hurt overlap significantly, the marginal returns to appropri-
its bargaining position in the alliance. Therefore, if ating partners’ knowledge may be decreasing. In this
Vistakon could commit not to take advantage of Nypro, case, although knowledge sharing in a single alliance
the latter would be more likely to share knowledge. may be unfeasible, we show that knowledge sharing
Our focus on commitments to encourage knowl- in a portfolio (with both A and B) can be feasible.
edge sharing yields recommendations that are often A second implication of our analysis that a focal
very different from and sometimes diametrically op- firm F is often better off by choosing partners that are
posed to what some strategic management approaches neither too similar nor too dissimilar to each other.
suggest. A key tenet of the learning race literature This is because partner substitutability (in both imple-
(Hamel 1991, Yan and Gray 1994, Inkpen and Beamish mentation and rival benefits) affects in opposite ways
1997, Khanna et al. 1998) is that alliance partners the probability of knowledge sharing and the share
should “maximize their receptivity to the knowledge of value captured by the focal firm. Thus, as suggested
and skills of their partner while limiting the trans- by some recent studies (Swaminathan and Moorman
parency of their own operations” (Mowery et al. 2009, Vasudeva and Anand 2011), the relationship
2002, p. 298). In our framework, firms engage in between alliance value and partner substitutability
“learning races” in some cases but exhibit “altruistic” may first increase and then decline.
behaviors in other cases. This may help explain some This paper makes two main contributions. First,
apparently puzzling behaviors, such as Toyota’s will- by emphasizing the need to encourage knowledge
ingness to teach lean manufacturing to GM, a com- sharing, we contribute toward putting the literature
petitor (Inkpen 2005). It may also help explain why on learning alliances on a more solid foundation.
sometimes firms such as Intel, Toyota, and Cisco Scholars have noted that many of the learning race
appear to deliberately reduce their own learning view’s recommendations suffer from a failure to rec-
capability through internal “Chinese walls” (Dyer ognize that the processes of value creation and value
and Nobeoka 2000, Gawer and Cusumano 2002, appropriation are inextricably linked (e.g., Zeng and
Steinhilber 2008).1 Hennart 2002). Some have also argued that the notion
Leading firms may also commit to low appropri- of a race to learn is “largely unrealistic” for it is un-
ation of a partner’s knowledge by creating alliance clear what would motivate a likely loser to join the
portfolios with desirable characteristics. Existing re- race (Inkpen 2002, p. 272). This paper incorporates a
search suggests that synergies between partners en- knowledge-sharing constraint into a model of learning
courage collaboration although partner substitutability in alliances and shows that its inclusion has important
hinders it by exacerbating competitive tensions (McEvily consequences for how alliances should be managed. In
et al. 2000, Bae and Gargiulo 2004, Lavie 2007). How- particular, the model may help explain why, although
ever, empirical work on the effects of partner sub- learning is an important goal in many alliances, only
stitutability or similarity has yielded largely inconclu- few firms actually appear to have a racing intent
sive results (Goerzen and Beamish 2005, Swaminathan (Mowery et al. 1996, Hennart et al. 1999, Inkpen 2000).
and Moorman 2009, Vasudeva and Anand 2011, Cui The second main contribution of the paper is to
and O’Connor 2012, Cui, 2013). reorient attention in strategy and economics from the
We provide a potential explanation for these mixed problem of how firms can protect their intellectual
results by distinguishing between two types of part- assets (e.g., Cohen et al. 2000, Katila et al. 2008) to the
ner substitutability: substitutability in implementa- problem of how leading firms can promote the health
tion and substitutability in rival benefits. Substitut- of their innovation ecosystems. Iansiti and Levien
ability in implementation refers to situations in which (2004) distinguish between “keystone organizations,”
the contribution of a former partner A to the imple- which share the benefits from collaboration with
mentation of a project can partly be replaced by the their partners, and “physical dominators,” which fo-
contribution of a staying partner B. Because this type cus on value appropriation. We examine how key-
of substitutability makes it easier for the focal firm, F, stone organizations can use commitment to help
to first appropriate A’s knowledge and then terminate create ecosystems conducive to knowledge sharing
the alliance with A, A’s incentives to share knowledge and innovation.
Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS 3

1.1. Related Literature conceptualized as commitments. These structures em-

This paper is related to several strands of the litera- body complex administrative structures that enhance
ture on alliances. By emphasizing contractual in- control over technology flows and facilitate collab-
completeness and ex ante investment incentives, our oration (Pisano 1989, Oxley and Wada 2009). In
model builds on the property-rights view of the firm. Section 5, we conclude that, although these contrac-
This literature studies how property rights should be tual solutions ameliorate the problem of shifting
allocated when concerns about holdup reduce part- bargaining positions, they only ensure efficiency
ners’ investment incentives (Grossman and Hart 1986, under stringent conditions. 5 Thus, organizational
Hart and Moore 1990). A key insight of this literature commitments remain important.6
is that, to maximize ex ante efficiency, property rights We focus on learning alliances and investments in
should be assigned to the party whose investment knowledge sharing. Our analysis is motivated by the
matters most to joint value creation. Aghion and observation that knowledge sharing often alters the
Tirole (1994), however, point out that an optimal partners’ bargaining positions through knowledge
allocation of property rights may not be achieved if spillovers. Information leakage and knowledge mis-
one of the partners is cash constrained. Panico (2017) appropriation are often cited as a major drawbacks of
argues that ownership structures that give partners R&D collaborations (Veugelers and Kesteloot 1994,
strong incentives to create value may also give them Arora and Merges 2004, Katila et al. 2008).7 Knowl-
strong incentives to engage in value appropriation. edge sharing can also shift the balance of power in
He studies how different ownership structures affect favor of the alliance partner that learns the fastest,
value creation and appropriation when partners in- potentially giving rise to “learning races” and in-
teract only once or repeatedly. This paper develops a stabilities (Hamel 1991, Yan and Gray 1994, Inkpen
model in which partners’ investments are equally and Beamish 1997).
costly and important; hence, an equitable division of The literature has largely focused on how firms can
the gains from collaboration tends to promote in- protect themselves against their knowledge spill-
vestment.2 The problem we emphasize is that in- overs in alliances. Katila et al. (2008), for instance, find
vestment may alter the partners’ bargaining positions that start-ups are more likely to enter corporate in-
(i.e., their outside options).3 vestment relationships with incumbents when they
A few papers examine how shifts in bargaining can defend themselves against misappropriation. De-
positions affect investment incentives. Nicita and fense mechanisms include not only patents and secrecy
Sepe (2012) note that, when investment affects out- (Cohen et al. 2000), but also the selection of partners
side options, overinvestment as well as underin- with limited opportunistic intent (Khanna et al. 1998,
vestment can occur. de Fontenay and Gans (2005) and Baum et al. 2000, Dussauge et al. 2000). How leading
Herbst and Walz (2017) explore in different settings firms can use commitments to encourage knowledge
the idea that bargaining with other potential part- sharing within their ecosystems has been much less
ners determines each firm’s outside option. We focus extensively studied. Our contribution is to show that
on organizational commitments to encourage invest- incorporating ex ante knowledge-sharing consider-
ment. The commitments we emphasize are constraints to ations in a model of learning alliances has important
one’s own absorptive capacity and alliance portfolio implications for alliance management.
design. Panico (2017) examines incentives to invest Finally, our alliance portfolio design analysis is
in value appropriation, which could be interpreted related to a growing literature examining how net-
as investments in absorptive capacity. However, in work structure affects knowledge sharing (Hansen
Panico’s (2017) model, investments in value appro- 2002, Argote et al. 2003, Reagans and McEvily 2003).
priation and investments in value capture are chosen The similarity of knowledge sources within a network
simultaneously. Commitment, on the other hand, is often regarded as an obstacle to effective knowl-
requires a sequential choice.4 edge sharing because it both reduces knowledge
The commitments that have been more extensively variety and creates competitive tensions (e.g., Baum
studied in economics are contractual or equity-based et al. 2000). However, empirical findings have been
(e.g., Williamson 1983, Oxley 1997). Williamson (1983) mixed (e.g., Goerzen and Beamish 2005, Swaminathan
considers a model in which producers can make and Moorman 2009, Cui 2013). Our work may help
relationship-specific investments, but such invest- reconcile these mixed findings by highlighting a type
ments create the potential for holdup. He shows that, of knowledge redundancy (substitutability in rival
under some conditions, efficient contracting can be benefits) that can actually facilitate knowledge sharing.
achieved if the buyer posts a “hostage” (such as a The rest of the paper is organized as follows. The basic
termination fee) that is delivered to the producer if model is presented in Section 2. There, we show that
the order is cancelled. Equity alliances (those involv- shifts in bargaining positions can destabilize alliances.
ing the sharing or exchange of equity) can also be Section 2.1 examines how uncertainty and different
 Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
4 Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS

types of uncertainty resolution affect knowledge then they must bargain over how to divide V. In the
sharing. Subsequent sections focus on commitments to case of Nypro and Vistakon, implementation means
facilitate knowledge sharing. Specifically, we exam- developing new disposable contact lenses together
ine investments in absorptive capacity (Section 3), and sharing the profits from this new product. If,
alliance portfolio design (Section 4), and contracts instead, F and A terminate their collaboration pre-
(Section 5). Section 6 concludes. Proofs are relegated maturely (before implementation), then F gets πF
to the online appendix. and A gets πA . We allow πF and πA to differ from zero
(the baseline payoffs in the absence of knowledge
2. Model sharing) because knowledge sharing may affect the
There are two firms, F and A, that participate in an firms’ competitive positions. For instance, πF and πA
alliance or collaboration. We sometimes refer to F as may both be positive if both F and A learn something
the focal firm because it may be involved in more than useful from their interaction that allows them to
one alliance. improve their existing products though, of course,
In period 0 (the commitment stage), the firms take one partner may learn more than the other (e.g.,
actions that influence how their payoffs in future πF > πA ). However, it could also be that one partner is
periods will evolve. We postpone the discussion of “expropriated” by the other (e.g., πF > 0, πA ≤ 0). This
these actions (which, in some cases, can be interpreted is the “swimming with sharks” or “information
as commitments) to Sections 3–5. For now, we assume leakage” scenario highlighted by many scholars (e.g.,
that no actions or commitments are available in pe- Veugelers and Kesteloot 1994, Katila et al. 2008).
riod 0. In the Nypro–Vistakon example, the risk that un-
In period 1 (the knowledge-sharing stage), the firms wanted knowledge spillovers may help Vistakon
simultaneously and noncooperatively choose whether develop alternative technologies to making lenses
to invest in knowledge sharing. We assume that these implies that πF > 0 (here, F stands for Vistakon and A
investments are not contractible. For instance, in the stands for Nypro).9 Similarly, the concern that Vista-
Nypro–Vistakon example discussed in the introduc- kon’s better understanding of Nypro’s manufacturing
tion, investments in knowledge sharing would in- process could constrain Nypro’s pricing flexibility
volve a more extensive integration between the two and bargaining power in future supply contracts
firms’ engineering teams than actually carried out. implies that πF > 0 or πA < 0. As we see, even if the
We assume that, if the firms invest in knowledge partners end up jointly implementing the innovation,
sharing, then each of them incurs a private cost 12 I > 0. the ex post outside options πF and πA affect how F
Thus, I denotes the total cost associated with knowl- and A share their joint value V.
edge sharing.8 Note that πF and πA only accrue if the collaboration
If F and A do not invest, then knowledge-sharing is prematurely terminated. Indeed, to the extent that
costs are not incurred, and F and A obtain their firm F can use the knowledge it gains from the alliance
“baseline” payoffs, which we normalize to zero. These without jeopardizing the alliance itself, these payoffs
payoffs may correspond to either if the alliance is not would simply be added to both V and πF , making the
formed or when an alliance is in place, but appropri- collaboration more attractive to F without making it
ability hazards prevent high levels of knowledge shar- less attractive to A. Our interest, on the other hand, is
ing as in the Nypro–Vistakon example. Thus, a payoff in “rival benefits,” which can potentially jeopardize
of zero may describe a situation in which F and A in- the collaboration by making it less attractive to the
troduce a new product together, but the new product other partner. In terms of our example, Vistakon
is not valuable because of limited knowledge shar- would only develop alternative technologies to mak-
ing. If both firms invest in knowledge sharing, then ing lenses if its collaboration with Nypro was termi-
payoffs V, πF , and πA are realized, where V is the nated. This outside option (worth πF ) would only be
value of the alliance and πF and πA are, respectively, F used if its joint project (worth V) were not completed.
and A’s (ex post) outside options. We assume that there Nevertheless, that Vistakon could terminate the col-
is no uncertainty about V, πF , and πA . In Section 2.1, laboration and still develop an alternative technol-
however, we introduce uncertainty and the cases of ogy influenced how the value of the joint project V
both symmetric resolution of uncertainty (both firms was divided.
learn the same information) and asymmetric resolution The key assumptions we make are that (i) it is
of uncertainty (the firms learn different information) impossible to contract on knowledge sharing and
in period 2 (the uncertainty resolution stage). that (ii) the firms cannot commit ex ante (before
In period 3 (the negotiation stage), the firms choose knowledge sharing) to implement the project to-
whether to continue to work together and earn V or gether ex post. These assumptions imply that the
exit and obtain their outside options. If they continue firms cannot be forced to collaborate; neither in the
to work together (i.e., they implement their project), initial phase of knowledge sharing nor in the subsequent
Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS 5

implementation phase. This could be because it is im- We refer to (2) as i’s knowledge-sharing constraint. If
possible for a court to verify, for instance, that a firm has this constraint holds, we say that knowledge sharing
done its best to transmit its knowledge or that a partner is privately profitable for firm i (assuming that the
is performing in a consummate rather than perfunctory other firm also shares knowledge). After some ma-
fashion during codevelopment. In countries with less nipulations, F’s and A’s knowledge-sharing con-
developed institutions, these assumptions may also straints can be rewritten as
reflect very high costs of using the legal system.
Note that firms do not unilaterally share knowl- V − I ≥ πA − πF , (3)
edge. We assume that, if F shares knowledge with A V − I ≥ πF − πA . (4)
but A does not, then F pays the knowledge-sharing
Unsurprisingly, F and A are more likely to collaborate
cost 12 I and A enjoys a benefit πuA ≥ 0 (the superscript u
if knowledge sharing creates substantial value (V − I
stands for unilateral knowledge sharing). If A shares
large). However, partners also care about their rela-
knowledge but F does not, then A pays the cost 12 I
tive bargaining positions. If πA > πF , then knowledge
and F enjoys a benefit πuF ≥ 0. Because knowledge
sharing shifts bargaining power in favor of A (that
sharing is not contractible, neither F nor A unilaterally
is, outside options shift from (0, 0) to (πF , πA ) with
shares its knowledge. Figure 1 summarizes the se-
πA > πF ), which makes F less likely to share its knowl-
quence of events.
edge (i.e., condition (3) is less likely to hold). Con-
In period 3, if the firms share knowledge and im-
versely, if πF > πA , then bargaining power shifts in
plement their project together, they must bargain
favor of F, and A is less likely to share knowledge
over the division of the surplus. We posit V ≥ πF + πA .
with F. Conditions (3) and (4) make clear that both
That is, after knowledge sharing, it is efficient for F
firms invest only if V − I ≥ |πF − πA |.10 Note that the
and A to implement their project together. This could
shifts in bargaining positions as measured by |πF − πA |
be because the partners have complementary capa-
only influence how the surplus generated by the al-
bilities in developing and commercializing the new
liance is divided between F and A. The condition for
product or because competition following a breakup
the efficiency of knowledge sharing, V > I, does not
dissipates rents. We assume bargaining is efficient and
depend on the relative magnitudes of πF and πA .
determined according to the Nash solution with equal
Because V − I ≥ |πF − πA | is stronger than V > I as
weights. Thus, the partners’ payoffs are given by
typical in property-rights models, the equilibrium can
1 1 exhibit underinvestment.
Πi  πi + [V − πF − πA ] − I, i  F, A. (1) We summarize this discussion as follows:
2 2
Proposition 1. Knowledge sharing is efficient when V ≥ I.
That is, each firm gets its ex post (after knowledge However, it is privately profitable for both firms only when
sharing) outside option πi plus half of the surplus
from implementation V − πF − πA minus the costs of V − I ≥ |πF − πA |. (5)
knowledge sharing 12 I. Knowledge sharing is efficient
if V ≥ I. Proposition 1 captures the idea that some value-
In a noncooperative equilibrium, knowledge shar- creating investments may not be undertaken when
ing occurs only if each firm i  F, A obtains more when knowledge sharing creates large shifts in bargaining
they both share knowledge than when they do not positions. Value creation is important but asymme-
share knowledge: tries in the evolution of outside options create a
wedge between equilibrium and socially efficient
Πi ≥ 0. (2) outcomes. Preserving the balance of power within the

Figure 1. Timing of the Model

 Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
6 Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS

alliance (a low πF − πA in absolute value) helps reduce synergies and asymmetric resolution of uncertainty
the risk of underinvestment. about outside options. We find that, contrary to
By emphasizing underinvestment in knowledge conventional wisdom, uncertainty often promotes
sharing rather than the failure of an alliance to be knowledge sharing in our setting.
formed, we implicitly assume that, despite underin-
vestment, the partners may still remain together as 2.1.1. Symmetric Resolution of Uncertainty About Syner-
was the case in the Nypro–Vistakon example. The gies. We begin with the case in which there is initially
SEMATECH alliance is another case in point. As uncertainty about the value of the synergies V. In
Grindley et al. (1994, p. 730) write, SEMATECH was period 1 (the knowledge-sharing stage), F and A only
know that V is distributed according to cumulative
[o]riginally intended to provide a research facility for
member firms to collaborate on projects to improve distribution function G over support [V, V]. However,
their semiconductor manufacturing process technol- in period 2 (after knowledge sharing but before
ogy. [However, . . . p]rocess technology expertise is implementation), both firms learn the realization of V.
central to the competitive advantage of individual Thus, there is symmetric information in period 3 (the
semiconductor manufacturers, and member firms were negotiating stage). We assume V > πF + πA > V so
reluctant to share such sensitive information. The so- that, if synergies V are low, it is efficient to terminate the
phistication of the manufacturing technology of SEM- alliance after knowledge sharing. Firms are risk neutral.
ATECH member firms also differed considerably, raising All other features of the model remain the same.
the danger that some firms could “free ride” on the In this setting, it is optimal for F to share knowledge
contributions of technology leaders. (conditional on A sharing knowledge) if
In response to these controversies, SEMATECH altered ∫
its research agenda. . . . The new research agenda has 1 V 1
πF + (V − πF − πA ) dG ≥ I. (6)
shifted the consortium’s focus from the development of 2 πF +πA 2
a complete state-of-the-art production process in its Austin
facility to knowledge diffusion and technology transfer. Similarly, it is optimal for A to share knowledge
(conditional on F sharing knowledge) if
SEMATECH provides another example in which
appropriability hazards and competitive tensions among ∫
1 V 1
member firms prevented participants from fully shar- πA + (V − πF − πA ) dG ≥ I. (7)
2 πF +πA 2
ing their knowledge. However, appropriability hazards
did not lead to the termination of the alliance. Instead, Intuitively, if firms share their knowledge, they ap-
the goal of the alliance was altered to focus on areas propriate their outside options (πF or πA ) plus half of
in which competitive tensions were less severe. the surplus, which accrues only when synergies are
In addition to noncontractible investment, several sufficiently high (V ≥ πF + πA ). By contrast, the con-
other features of the model are also well suited for the dition for efficient knowledge sharing is
analysis of knowledge sharing. In our model, part- ∫ V
ners’ investments are highly complementary (V is πF + πA + (V − πF − πA )dG ≥ I. (8)
only created if both partners invest). Our results re- πF +πA
main valid for arbitrarily small investment costs. 11
A firm’s investment may affect not only its own Proposition 2 characterizes the equilibrium under
outside option, but also the outside option of its synergy uncertainty and shows how greater uncer-
partner. It is difficult to think of many investments tainty (modeled as a mean preserving spread of the
not involving knowledge sharing when all these as- original distribution G) affects knowledge sharing.
sumptions are verified. Also, the next section con- Proposition 2.
siders investments in absorptive capacity. There, we i. The equilibrium can exhibit underinvestment in knowl-
show that a “strong” partner may invest resources to edge sharing.
restrict its own absorptive capacity, thus lowering the ii. Early termination of an alliance (before implementa-
value of its outside option. In the case of more tra- tion) occurs with positive probability but is always efficient.
ditional investments in physical or human capital, we iii. An increase in uncertainty about synergies makes
would instead expect investment to improve one’s investment in knowledge sharing more likely.
own outside options.
As in the model without uncertainty, the nonco-
operative equilibrium can exhibit underinvestment in
2.1. Uncertainty knowledge sharing. However, unlike the baseline
This subsection explores how uncertainty affects firms’ model, now, an alliance can be terminated early—
incentives to share knowledge. We consider two after knowledge sharing but before implementation—
scenarios: symmetric resolution of uncertainty about if synergies turn out to be low. Proposition 2 also
Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS 7

shows that the incentives for knowledge sharing are Proposition 3 shows that, although asymmetric in-
greater when there is more uncertainty about the formation always creates an inefficiency relative to
value of synergies. Intuitively, a mean preserving the first best, compared with a benchmark scenario
spread of the distribution of V increases the likelihood with complete information, asymmetric information
of extreme (very low and very high) realizations of V. can improve welfare. The intuition is as follows.
If V is very low, after knowledge sharing, the partners When making an offer x at the negotiating stage, the
can still obtain πF and πA . The downside is, therefore, uninformed player (firm F) trades off a smaller gain
limited. On the other hand, if V is very high, the V − x when the offer is accepted for a greater prob-
partners can share this large value. This upside of ability Pr(πA ≤ x) that the offer is accepted. In the
knowledge sharing increases as more extreme reali- optimum, therefore, F makes an offer that is rejected
zations of V become possible. In short, there is an with positive probability. This is inefficient, and welfare
option value associated with knowledge sharing, and is reduced.
the value of that option increases with the level of However, the incentives to share knowledge can
uncertainty over synergies. be higher in the one-sided asymmetric information
scenario than in the benchmark scenario. The reason
2.1.2. Asymmetric Resolution of Uncertainty About is that asymmetric information improves the bar-
Outside Options. Next, we consider a scenario in gaining position of the informed party (A). Knowl-
which there is initially uncertainty about one of the edge sharing may not be possible with symmetric
partners’ (ex post) outside options, namely πA . In information (because πF is much larger than πA ), but
period 1 (the knowledge-sharing stage), F and A only asymmetric information can improve A’s bargaining
know that πA is distributed according to some dis- position so much that knowledge sharing is possible
tribution function. In period 2 (after knowledge shar- under one-sided asymmetric information. Thus, asym-
ing but before implementation), A learns the realiza- metric information can help balance an initially un-
tion of πA , but F learns nothing. Thus, in period 3 (the balanced relationship.
negotiating stage), there is one-sided asymmetric
information (with which A has an informational ad- 3. Commitments to Encourage
vantage). For tractability, we assume that πA is dis- Knowledge Sharing
tributed uniformly on [0, V − πF ]. Note that, because So far, we have shown that appropriability hazards
V ≥ πF + πA for all realizations of πA , premature ter- can lead to underinvestment in knowledge sharing.
mination of the alliance is never efficient. Firms are Next, we consider commitments that prospective
risk neutral. partners can make to encourage knowledge sharing
The project is implemented if the firms agree on a when appropriability hazards are severe. These strat-
division of V. Bargaining occurs as follows. With egies include (i) restricting one’s own absorptive ca-
probability 1/2, F makes a take-it-or-leave-it offer pacity, (ii) alliance portfolio design, and (iii) con-
to A, and with probability 1/2, A makes a take-it-or- tractual solutions. Throughout, we assume with no
leave-it offer to F. This procedure, if πA is known to essential loss of generality that F is the “strong”
both partners, would produce exactly the same out- partner in the sense that πF > πA . Thus, the binding
come (in expectation) as the Nash bargaining solution knowledge-sharing constraint is generally that of the
in Section 2. All other features of the model remain “weak” partner A.13
the same.
We compare this one-sided asymmetric informa-
3.1. Restricting One’s Own Absorptive Capacity
tion scenario to a benchmark scenario in which πA is
An influential literature in strategy stresses that at-
nonstochastic and given by (V − πF )/2 (the expected
tempts to appropriate the returns from collaboration
value of πA when πA is distributed uniformly on
may generate learning races, in which partners try to
[0, V − πF ]). Welfare is defined as the sum of the firms’
absorb their partners’ knowledge while protecting
expected payoffs in period 1 (before the resolution of
their own (e.g., Hamel 1991, Khanna et al. 1998). Our
uncertainty). We can show the following:
model suggests that strengthening one’s own bar-
Proposition 3. gaining position beyond a point may be counterpro-
i. In the one-sided asymmetric information scenario, ductive because it may discourage knowledge sharing.
premature termination of the alliance occurs with positive In this section, we develop a simple extension of the
probability and is inefficient. In the benchmark scenario, model in which firms invest in their ability to absorb
premature termination of the alliance never occurs. external knowledge, which, in turn, improves their
ii. Despite this inefficiency, investments in knowledge bargaining position vis-à-vis their partner. We show
sharing and welfare can be higher in the one-sided asymmetric that, when the need of encouraging knowledge shar-
information scenario than in the benchmark scenario.12 ing is taken into account, firms do not always want to
 Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
8 Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS

maximize their receptivity to their partners’ knowl- Proposition 4 shows that, when firms are very
edge. Instead, they may sometimes intentionally re- asymmetric in terms of their absorptive capacities
strict their own learning capability to encourage (case (i)), knowledge sharing does not occur. Asym-
knowledge sharing. Thus, the model can potentially metries lead to a breakdown of cooperation, and the
explain why, in reality, firms seldom appear to exhibit resulting outcome is inefficient. If, instead, firms are
a racing intent (Mowery et al. 1996, Hennart et al. relatively similar in terms of their absorptive capac-
1999, Inkpen 2000) and their behavior is instead best ities (case (iii)), a learning race takes place. Both firms
described as cooperative (Inkpen 2005). maximally invest in absorptive capacity, but this still
We assume that, in period 0 (prior to the knowledge- leads to a fairly symmetric outcome. Thus, they still find
sharing stage), firms invest in their absorptive capacity. it privately profitable to invest in knowledge sharing.
These investments are irreversible and observable and The most interesting case arises when absorptive
affect firms’ outside options in case of premature capacities are asymmetric but not excessively so
termination of the alliance.14 Specifically, we assume (case (ii)). A strategy of maximizing absorptive ca-
that, if firm i  F, A chooses a low absorptive capacity pacity becomes self-defeating for the stronger part-
in period 0 (at cost cLi ), then its ex post outside option ner F. By maximizing its absorptive capacity, firm F
in period 2 (following knowledge sharing and pre- discourages firm A from sharing its knowledge. F is
mature termination of the alliance) is πLi . Con- better off by limiting its absorptive capacity to πLF ,
versely, if firm i chooses a high absorptive capacity thus sharing more surplus with A.
(at cost cHi ), then, in period 2, its ex post outside op- There are many examples of firms that appear to
tion is πH i , where πi > πi and V ≥ πF + πA so that
H L H H
deliberately reduce their own absorptive capacity to
premature termination of the alliance is not efficient.15 facilitate cooperation with their partners. Cisco, for
The learning race literature suggests that F should instance, is a firm that has successfully managed a
select πH F and A should select πA . To bias the results
H
large number of alliances in a variety of sectors, ge-
in this direction, we assume that investments in ab- ographies, and technological areas. Cisco recognizes
sorptive capacity are equally costly (i.e., cH F  cF and
L
that conflicts can arise when partners are exposed to
cA  cA ). Thus, deviations from the outcome (πF , πH
H L H
A) each other’s knowledge. Steve Steinhilber (2008, pp.
must emerge not from cost considerations, but for 101, 114), vice president of strategic alliances at Cisco,
purely strategic reasons. We interpret the choice of notes that,
the strong partner F to restrict its own absorptive
One of the most contentious issues in negotiating the
capacity (i.e., to choose πLF instead of πH F ) as a com- confidentiality terms of an alliance agreement is the
mitment to encourage knowledge sharing. Finally, to
treatment of residuals—that is, general knowledge,
reduce the number of cases to consider, we assume know-how, and the skills that each partner’s em-
that πLF ≥ πH A . This implies that the focal firm F is al- ployees will gain by being exposed to the other
ways the “faster learner” (or stronger partner) and A party’s confidential information . . . you face consider-
is always the “slower learner” (or weaker partner). able risk. . . . You could open your doors to a company
All other features of the model in Section 2 remain that could hurt you in your own market over time, gain
the same. competitive advantages, or acquire unique knowledge
Proposition 4 shows that the learning race intuition or skills that it could not have obtained otherwise.
is, in general, incorrect. Partners do sometimes en-
Steinhilber recommends that partners establish ground
gage in learning races and maximize their absorptive
rules to manage information security and intellectual
capacity (case (iii) as follows). However, in some cir-
property rights. These rules should be designed not
cumstances, they purposefully limit their absorptive
simply to protect one’s own resources, but also to
capacity and increase their partner’s payoff (case (ii)).
ensure that all the partners are treated fairly and
Proposition 4. Suppose knowledge sharing is efficient (V ≥ I), nobody’s knowledge is mishandled. In particular,
firm F is always the faster learner (πLF ≥ πH A ), and invest- Steinhilber (2008, p. 119, emphasis added) suggests
ments in absorptive capacity are equally costly (cH F  cF and
L that firewalls may sometimes be created to prevent
cA  cA ). In equilibrium,
H L Cisco from learning too much from its partners.
i. If V − I < πLF − πHA , the firms do not invest in knowl- Specifically, he recommends
edge sharing. The choices of absorptive capacity are [s]etting clear parameters in your agreements that
inconsequential. identify the information to be shared and the per-
ii. If V − I ∈ [πLF − πHA , πF − πA ), the firms invest in
H H
mitted use of such information. In certain instances, it
knowledge sharing. Firm F selects πLF , and firm A selects πHA. may be necessary to restrict information to some
iii. If V − I ≥ πHF − π H
A , the firms invest in knowledge employees and to set up firewalls to prevent tainting
sharing. Firm F selects πH F , and firm A selects πA .
H
other groups within the company that are developing
Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS 9

similar technology independently. . . . Setting up train- responsibility, and no products. . . . And IAL has achieved
ing and procedures to protect your partner’s confidential an extra measure of credibility. It comes, first, from the
information and watching for actions by your partner fact that they are very good, and second, that they are
that may signal an improper use of your own information. not in a business” (quoted in Gawer and Cusumano
In the context of our model, we can interpret Cisco’s 2002, p. 128, emphasis in original).
internal Chinese walls as a commitment to reduce its IAL head Craig Kinnie (also quoted in Gawer and
own absorptive capacity. If the relationship between Cusumano 2002, p. 128) similarly notes the impor-
Cisco and one of its partners were to break down, tance of IAL being perceived as fair and impartial.
Cisco would arguably see no impediment to re- Consistent with the literature on credible commit-
moving these barriers and using any information that ments and our model, he also emphasizes that the
it may have gleaned during the collaboration. How- separation of roles between IAL and other Intel units
ever, this information would only be transmitted with must be externally visible:
delay after the alliance has been terminated and the To deserve the trust, a separation of roles inside the
communication barriers have been removed. These organization is an absolute necessity. And it has to be
delays would imply lower rival benefits πF to Cisco and observable from the outside. There is a tension be-
higher rival benefits πA to its partner (because com- tween the two perspectives: the Laboratory perspec-
petition between Cisco and the partner would be tive and the product groups’ perspective. . . . It is also
important to keep that polarity for the sake of enrolling
delayed). Both effects would encourage Cisco’s part-
others on the outside. If the labs were buried inside of
ners to invest in knowledge sharing. the product groups, the perceived neutrality and
Toyota and Intel provide further illustrations of trustworthiness of the labs would go down.
how setting up internal Chinese walls can facilitate
collaboration. Toyota benefits greatly from the de- Taken together, these examples suggest that com-
velopment and diffusion of production knowledge mitments to protect a partner’s confidential infor-
within its supplier network. To encourage partners to mation can help encourage knowledge sharing in
share their knowledge, Toyota made the conscious alliances. Interestingly, and in line with our model, in
decision to separate its operations management con- all these examples it was the leading firm (Cisco, Intel,
sulting division from its purchasing division, “so that Toyota) that reduced its own absorptive capacity.
suppliers can work with the consultants without Clearly, incumbents do not always behave like
fearing that purchasing will ask for a price decrease “sharks” but sometimes adopt apparently altruistic
after the consultation” (Dyer and Nobeoka 2000, pp. behaviors, motivated by the need to engender trust
358–359). As Hajime Ohba, the head of the Toyota and encourage knowledge sharing.16
Supplier Support Center, notes, “Our job is to help
4. Alliance Portfolio Design
suppliers improve, not to worry about who gets the
In this section, we study how alliance portfolios
additional profits” (quoted in Dyer and Nobeoka can be designed to facilitate knowledge sharing. We
2000, p. 359). Because suppliers appropriate in the assume that, in period 0, the focal firm F chooses
short run a large fraction of the benefits from collabo- (i) whether to be involved in one or more dyadic
ration, this arrangement gives them strong incentives alliances and, if it is involved in more than one alli-
to participate in knowledge-sharing activities. ance, (ii) the degree of similarity among its alliance
Intel also faces significant tensions when collabo- partners. The key contribution of this section is to
rating with partners (Gawer and Cusumano 2002). On distinguish between two types of substitutability and
the one hand, Intel wants to expand demand for its show that there is a type of substitutability that fa-
microprocessors (“job 1”), which requires encour- cilitates, not hinders, knowledge sharing. We also
aging the entry of complementary products, such as show that, regardless of which type of substitutability
new graphics cards and chip sets. On the other hand, is more important, F typically benefits from choosing
Intel also wants to grow profitable businesses in partners that are neither too similar nor too dissimilar
complementary markets (“job 2”). Intel’s dilemma is from each other. A key message of this section is that,
how to encourage “complementors” to enter although because there is a “good” type of substitutability,
ex post it may have an incentive to compete with alliance portfolios can facilitate knowledge sharing.
them. Gawer and Cusumano (2002) argue that one Knowledge sharing may not be feasible in a single
method is to create separate groups or divisions alliance but may be feasible in a portfolio.
pursuing job 1 or 2. For example, the Intel Architec-
ture Laboratory (IAL) was explicitly structured as 4.1. Partner Substitutability and Knowledge Sharing
a cost center and rewarded for its success in pro- A recent literature focuses on alliance portfolios and
moting the health of the whole ecosystem (job 1). interdependencies among partners. At the risk of
As Andy Grove notes, “IAL has no profit-and-loss simplifying a complex subject, this literature suggests
 Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
10 Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS

that overlaps among alliance partners reduce value: We incorporate these two notions of substitut-
they reduce the potential for synergies and increase ability in a simple extension of the model studied in
conflict among partners. The result, according to this Section 2. To reduce notation, we assume that F’s
logic, is lower stability of alliances and lower value alliances are completely symmetric in terms of pay-
to the focal firm (e.g., Vasudeva and Anand 2011, offs. Thus, for instance, F obtains the same baseline
Wassmer and Dussauge 2011).17 outside option (or rival benefit) πF regardless of
We revisit these ideas through the lens of our whether the alliance with A or B is prematurely ter-
model. We consider a setting in which a focal firm F minated. If neither pair shares knowledge, all firms
collaborates with two partners, A and B. Thus, there obtain the same payoffs, normalized to zero. A’s
are two dyadic alliances, one between F and A and one and B’s outside options following premature termina-
between F and B. We say that A and B are substitutes tion are also set for simplicity equal to zero: πA  πB  0.
if they allow F to access similar pools of knowledge. We begin with the case in which F and A share
We distinguish between two types of substitutability: knowledge but F and B do not. Then, it is as if B did not
substitutability in implementation and substitutability exist and the analysis follows the same steps as in
in rival benefits. Section 2. Knowledge sharing is efficient if V ≥ I;
Substitutability in implementation refers to situa- however, it is privately profitable for both F and A
tions in which the contribution of a former partner to only if V − I ≥ πF . Thus, in equilibrium, there can be
the implementation of a project can partly be replaced underinvestment in knowledge sharing.
by the contribution of the staying partner. For in- Next, consider the case in which knowledge shar-
stance, suppose F and A share knowledge to develop ing occurs in both alliances. Let v(S) be the value
new artificial intelligence (AI) algorithms. If this al- created by coalition S ∈ {F, A, B, FA, FB, AB, FAB} when
liance is prematurely terminated (after knowledge (i) the firms in S implement their projects together and
sharing but before implementation), F may be able to (ii) the alliances between the firms in S and the firms
develop its own algorithms (a “rival” project). In not in S are terminated prematurely. Thus, v(FAB) is
implementing this rival project, F’s other (staying) the value created by coalition FAB when both F and A
partner may help. Specifically, if the alliance be- and F and B implement their projects together. Value
tween B and F is not prematurely terminated, the v(FA) is created by coalition FA when F and A im-
value of the rival project increases from πF to πF + sImp . plement their project together but the alliance be-
Here, sImp ≥ 0 measures the value of B’s contribution tween F and B is prematurely terminated. Value v(A)
to the implementation of the rival project.18 is created by A when the alliance between F and A is
By contrast, substitutability in rival benefits refers prematurely terminated.19 All values v are gross of
to situations in which the knowledge gained by for- knowledge-sharing costs.
mer partners is partly overlapping so that the rival We assume v(FAB)  2V + b. V is the “baseline”
benefits from prematurely terminating both alliances value created by F and A (or F and B) working to-
are lower. For example, suppose F acquires AI com- gether. The synergistic benefit b ≥ 0 captures com-
petencies from both A and B and there are two rival plementarities between A and B in implementation.
projects in which F can redeploy these capabilities: By combining A’s and B’s capabilities during the
algorithms for image classification and algorithms for implementation stage, the coalition FAB creates b
speech recognition. Suppose further that algorithms more value than if A and B were involved in two
for image classification are more valuable to F than completely unrelated alliances. When the alliance
algorithms for speech recognition. If only one alliance between F and B is terminated prematurely, the co-
is prematurely terminated (say the one with A), then F alition FA obtains v(FA)  V + πF + sImp , where V is
could redeploy the engineers working with A to work the value created by F and A implementing their
on image classification, thus yielding rival benefits πF . project together and πF + sImp is the value created by F
However, if both alliances are prematurely termi- and A working on the rival project originating from
nated, both the engineers working with A and the the terminated alliance with B. We assume sImp ∈ [0,
engineers working with B could be redeployed. One V − πF ], which implies that prematurely terminating
group would work on image classification, yielding an alliance is not efficient.
rival benefits πF , and the other group would work on If both alliances are terminated prematurely, the
speech recognition, yielding rival benefits πF − sRB , values created by coalitions F, A, and B are, respec-
so that F’s total payoff would be 2πF − sRB . Here, tively v(F)  2πF − sRB , v(A)  0, and v(B)  0. Non-
sRB ≥ 0 measures substitutability in rival benefits. focal partners always get a payoff of zero if they do not
This notion captures the idea that, if A and B’s implement their projects with F. The focal partner F,
knowledge bases are similar, the marginal returns to by contrast, can work on two rival projects, earning
prematurely terminating multiple alliances are likely to 2πF − sRB , where sRB measures substitutability in rival
be decreasing. benefits. We assume sRB ∈ [0, πF ].20
Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS 11

Table 1 summarizes the values created by all the and B, on average, lose as a result of implementation
possible nonempty coalitions following knowledge substitutability, and F gains. Finally, substitutability
sharing. If the alliances between F and A and F and B in rival benefits sRB lowers the outside option of the
were completely unrelated, we would have b  sImp  focal firm F in case both alliances are terminated and,
sRB  0. therefore, weakens F’s bargaining position vis-à-vis A
Following Hart and Moore (1990), we use the and B. Thus, sRB appears with a negative sign in
Shapley value to assign payoffs to individual firms. If Equation (9) and with a positive sign in Equation (10).
knowledge sharing occurs, then the “grand coalition” Define ΠF  φF − I, ΠA  φA − 12 I, and ΠB  φB − 12 I.
FAB emerges because it creates the greatest total Knowledge sharing in both alliances is privately
value.21 The Shapley value assigns each firm its ex- profitable for all partners if Πi ≥ 0 for all i  F, A, B.
pected marginal contribution assuming that the or- Knowledge sharing in both alliances is efficient if
der in which they join the grand coalition is ran- 2V + b ≥ 2I. To facilitate the comparison with the
dom. Formally, for i  F, A, B and any subset of firms single alliance case, we assume that the knowledge-
S ⊆ N  {F, A, B}, the value assigned to firm i is sharing condition for the focal firm holds, that is,
ΠF ≥ 0, but those of the weak partners may not hold.
∑ |S|!(|N | − |S| − 1)!
φi  (v(S ∪ {i}) − v(S)), We have the following:
S⊆N\{i} |N |!
Proposition 5. Suppose V ≥ I. In alliance portfolios,
where |S| denotes the number of firms in S and, for i. Substitutability in rival benefits facilitates knowl-
any positive integer r, r!  1 × 2 × . . . × r, and 0!  1. edge sharing, and substitutability in implementation hin-
The Shapley value yields the expressions for the ders it. Synergies in implementation also facilitate knowl-
edge sharing.
firms’ (gross) payoffs:22
ii. Synergies in implementation have a larger effect in
1 1( ) facilitating knowledge sharing than substitutability in
φF  V + πF + b + sImp − sRB , (9) implementation has in hindering it. More precisely, sup-
3 3
1 1 1 ( ) pose b  k1 + λ and sImp  k2 + λ. Then, knowledge shar-
φA  φB  (V − πF ) + b − sImp − sRB . (10) ing becomes more likely in alliance portfolios as λ grows.
2 3 6
These results follow from the knowledge-sharing
As the single alliance case, the fact that F can use the
constraints of the weak partners:
knowledge learned from its partners on rival projects
shifts the balance of power in its favor; F gets an 2 1( )
additional 12 πF from each partner, and A and B lose the ΠA , ΠB ≥ 0 ⇔ V − I − πF + b − sImp − sRB ≥ 0.
3 3
same amount. The synergistic value b is created only if (11)
all the firms work together; hence, b is split equally
among them. Substitutability in implementation has The analysis suggests two reasons why substitut-
both costs and benefits for A and B. On the one hand, A ability may not be as bad at discouraging knowledge
can be replaced by B, which lowers A’s bargaining sharing as generally thought. First, we identify a
power vis-à-vis F. On the other hand, A may replace B. type of substitutability, substitutability in rival ben-
In this case, A shares the benefits of substitutability in efits sRB , which facilitates rather than hinders knowledge
implementation with F. The cost of lower bargaining sharing. Second, we show that synergies in imple-
power for A (and similarly for B) is 13 sImp , and the mentation b have a larger effect in facilitating knowl-
benefit of replacing B is just 16 sImp . The benefit is half edge sharing than substitutability in implementation
the cost because the gains from replacing B do not just sImp has in hindering it. The reason, as mentioned,
accrue to A but must be split between A and F. Thus, A is that substitutability in implementation has some

Table 1. Coalitional Values After Knowledge Sharing

Outcomes of negotiations Coalitional values

Neither collaboration is
v(FAB)  2V + b
terminated prematurely
One and only one v(FA)  v(FB)  V + πF + sImp
collaboration is terminated v(AB)  0
prematurely
Both collaborations are v(F)  2πF − sRB
terminated prematurely v(A)  v(B)  0
 Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
12 Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS

beneficial effects for nonfocal partners. A nonfocal iv. If α > 12 and V − I − πF + 23 b > 0, then F’s equilib-
partner may lose a lot when it is replaced, but it may rium payoff first rises then declines with Ψ. The opti-
also gain a little when it is instrumental in replacing mum level of similarity from F’s viewpoint is Ψ∗  (3(V −
another nonfocal partner. I − πF ) + 2b)/(2α − 1).23
An implication of our analysis is that knowledge
Proposition 7 suggests that the relationship be-
sharing in a portfolio can be feasible, and knowledge
tween the value that F can appropriate and the degree
sharing in a single alliance may not be.
of similarity of its alliance partners is typically non-
Proposition 6. Suppose V ≥ I. If V − I ∈ [πF − 23 b + 13 (sImp − monotonic. The reason is that there are, in general,
sRB ), πF ), then knowledge sharing in a portfolio is feasible, two conflicting effects at play: (i) on the focal firm’s
and knowledge sharing in a single alliance is not. payoff conditional on knowledge sharing and (ii) on the
Proposition 6 suggests that one way to encourage weak partners’ incentives to share knowledge.
knowledge sharing when appropriability hazards are Consider substitutability in implementation first.
present is to create alliance portfolios with desirable A high degree of similarity Ψ between nonfocal
characteristics. The strategy of adding a third partner partners benefits F conditional on knowledge sharing
(i.e., a second dyadic alliance) is beneficial not only because it strengthens F’s bargaining power. How-
when there are synergies in implementation between ever, too much similarity reduces the nonfocal part-
nonfocal partners, but also when substitutability in ners’ incentives to share knowledge with detrimental
rival benefits is significant. effects on F as well. Thus, the optimum level of
similarity between nonfocal partners from F’s view-
4.2. Similarity Between Alliance Partners point is intermediate (Proposition 7, case (iv)).
Besides selecting the “right” number of partners, a When substitutability is in rival benefits, these ef-
focal firm must also select partners with the right fects are reversed. Partner similarity harms F condi-
characteristics. Here, we examine the following ques- tional on knowledge sharing but also induces the non-
tion: should the focal firm F choose nonfocal partners focal partners to share knowledge. Initially (for low
that are similar to each other (i.e., with a large degree levels of similarity), knowledge sharing in an alliance
of substitutability) or partners that are different from portfolio is not sustainable, and F’s payoff is low.
each other (with a low degree of substitutability)? However, when Ψ crosses Ψ∗ , knowledge sharing
To answer this question, let sImp  αΨ and sRB  becomes sustainable, so F’s payoff suddenly rises. As Ψ
(1 − α)Ψ. The parameter Ψ ≥ 0 measures the degree of grows further, substitutability in rival benefits mono-
similarity between A and B. A high degree of simi- tonically reduces F’s payoff (Proposition 7, case (i)).
larity between A and B increases both substitutability Thus, regardless of which type of substitutability is
in implementation and in rival benefits. α ∈ [0, 1] is a more important, F is often better off by choosing
parameter capturing to what extent substitutability partners that are neither too similar nor too dissimilar
in implementation is more important than substi- to each other. The relationship between F’s payoff and
tutability in rival benefit. If α  1, then only substi- nonfocal partner similarity Ψ has often an inverted-U
tutability in implementation matters. If α  0, then shape (cases (i) and (iv)).
only substitutability in rival benefit is important. By This basic intuition is complicated by the fact that
varying α, we can examine how our results depend on initial conditions also matter. Depending on the initial
the type of substitutability. propensity of its alliance partners to share knowl-
Proposition 7 shows that, regardless of which type edge (whether V − I − πF + 23 b + 0), only part of the
of substitutability is more important, F should often inverted-U curve may be observed. Case (ii) describes
choose partners that are neither too similar or too a situation in which, because substitutability is
dissimilar from each other. mostly in rival benefits (α < 12) and knowledge sharing
can always be sustained, partner similarity is always
Proposition 7. Suppose sImp  αΨ and sRB  (1 − α)Ψ, detrimental to F. Hence, the optimal value of partner
where Ψ ≥ 0 and α ∈ [0, 1]. similarity from F’s viewpoint is zero. Case (iii) de-
i. If α < 12 and V − I − πF + 23 b ≤ 0, then F’s equilibrium scribes a situation in which knowledge sharing can
payoff first rises then declines with Ψ. The optimum level never be sustained. In that case, because there is no
of similarity from F’s viewpoint is Ψ∗  (3(V − I − πF ) + 2b)/ knowledge sharing, the degree of partner similarity
(2α − 1). is inconsequential.
ii. If α < 12 and V − I − πF + 23 b > 0, then F’s equilib- The complexity of these predictions may help ex-
rium payoff monotonically declines with Ψ. The optimum plain the variety of often conflicting empirical results.
level of similarity from F’s viewpoint is zero. Swaminathan and Moorman (2009) find that alliance
iii. If α > 12 and V − I − πF + 23 b ≤ 0, then F’s equilib- announcements create value (i.e., abnormal stock
rium payoff is independent of Ψ and always equal to zero. returns), especially when the degree to which the
Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS 13

firm’s network of alliances involves firms that pos- be less than A’s payoff from not reneging and sharing
sess nonredundant knowledge, skills, and capabil- knowledge, that is,
ities is moderate. Vasudeva and Anand (2011) find an
inverted U–shaped relationship between the techno- 1 1
x ≤ (V − πF + πA ) + x − I
logical diversity in a focal firm’s alliance portfolio and 2 2
the likelihood that the focal firm cites its partner’s ⇐⇒ (V − πF + πA ) − I ≥ 0. (13)
patents. Cui and O’Connor (2012) find no statistically
significant relationship between alliance portfolio But (13) cannot hold because we assumed V − πF +
resource diversity and innovation although several πA − I < 0. That is, a contract that encourages knowl-
interaction effects are significant, and Goerzen and edge sharing also encourages opportunistic behavior
Beamish (2005) and Cui (2013) find monotonic rela- by A.
tionships. Our theory can help rationalize these findings;
however, further research is surely needed. 5.3. Equity Alliances
In the model in Section 2, there is no contract speci-
5. Contractual Solutions fying how the returns from collaboration are divided.
Contractual solutions to the holdup problem have We simply assumed that, given outside options (πF , πA ),
been extensively studied in economics (see Segal and partners split the surplus V − (πF + πA ) equally among
Whinston (2013) for a survey). In this section, we themselves. However, if the partners can commit to a
examine contractual solutions that appear to be im- contract that specifies a certain division of the surplus
portant in practice. We argue that, although contracts should they implement their project together, they
can facilitate knowledge sharing, their usefulness is can improve outcomes. Consider a contract such that,
often limited. if the partners implement their project together, a
share λ ∈ [0, 1] of the surplus goes to A, and the share
5.1. Payments to Share Knowledge
1 − λ of the surplus goes to F. The partners’ payoffs
A contract specifying payments from the strong partner
conditional on knowledge sharing are24
to the weak partner to encourage knowledge sharing
does not solve the underinvestment problem high- 1
lighted in this paper. The reason is that knowledge ΠEF  πF + (1 − λ)[V − πF − πA ] − I, (14)
2
sharing is not observable by a court. Thus, the weak 1
partner would sign the contract, accept the payment, ΠA  πA + λ[V − πF − πA ] − I.
E
(15)
2
and then would not share knowledge if that was not in
its own interest. Importantly, contract λ does not oblige the firms to
share knowledge or implement the project together.
5.2. Termination Fee Contracts Instead, F and A can choose between doing so and
Termination fee contracts may also appear to be a realizing the payoffs specified under the contract or
potential solution (Williamson 1983). However, they separating and obtaining their respective outside
do not work in our model. options. The key assumption we make here is that the
Suppose collaboration is efficient (V ≥ I) but the partners are unable to renegotiate the contract λ in
weak partner will not share knowledge (i.e., V − πF + period 2 (after knowledge sharing). If they could do
πA < I). To avoid this problem, suppose F offers a so, then we would expect their (re)negotiation to
contract to A that pays x to A if the collaboration produce the same outcome as in Section 2 because
terminates prematurely before joint implementation. their ex ante bargaining weights are the same.
In this case, once knowledge sharing has occurred, F’s Proposition 8 compares “equity” alliances associ-
outside option is πF − x, and A’s outside option is ated with contract λ to the “nonequity” alliances
πA + x. The new gross payoffs that accrue to F and A in which λ  1/2 (that is, the alliances studied in
after negotiations are, thus, respectively, 12 (V + πF − Section 2). We say that knowledge sharing can be
πA ) − x and 12 (V − πF + πA ) + x. implemented by an equity alliance if there exists an
For knowledge sharing to occur, it must be prof- equity share λ ∈ [0, 1] such that both partners are
itable for the weak partner: willing to share knowledge. Similarly, we say that
knowledge sharing can be implemented by a non-
1 1 equity alliance if both partners are willing to share
(V − πF + πA ) + x ≥ I. (12)
2 2 knowledge when λ  1/2.
However, with a termination fee, the weak partner Proposition 8.
may behave opportunistically. Specifically, A may i. If knowledge sharing can be implemented by a non-
not share knowledge simply to force termination and equity alliance, then it can also be implemented by an
get a payment of x. To prevent such behavior, x has to equity alliance. However, the converse is not true.
 Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
14 Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS

ii. If knowledge sharing can be implemented by and appropriate a greater share of the collabora-
an equity alliance but not by a nonequity alliance, then tive pie. All these risks, if foreseen, can discourage
πF > πA implies λ > 1/2, and πF < πA implies λ < 1/2. knowledge sharing unless contractual or other types of
iii. Even if knowledge sharing is efficient, knowledge safeguards exist.
sharing may not be implementable by an equity alliance. Our paper emphasizes commitments that leading
Part (i) is obvious. Nonequity alliances are a special firms can make to encourage knowledge sharing. Our
case of equity alliances. Thus, if knowledge sharing theory does not negate the importance of safeguards
can be implemented by a nonequity alliance, it can or knowledge protection; however, it does provide
also be implemented by an equity alliance, but the a more nuanced perspective by incorporating also
reverse is not necessarily true. the incentives of leading firms to create a “safe space”
Part (ii) shows how equity alliances can be struc- for collaboration.
tured to facilitate knowledge sharing when nonequity The paper has several implications that are relevant
alliances are ineffective. Suppose nonequity alliances to managers and practitioners. A basic but important
are ineffective because knowledge sharing shifts bar- point is that managers must be aware of their part-
gaining positions excessively in favor of F (i.e., πF is ners’ incentives to collaborate. If leading firms do not
much larger than πA ). Then, F may grant the weaker leave sufficient rents to their alliance partners, the latter
partner A a larger share of the surplus (λ > 1/2) to will not participate in collaborative endeavors or will
encourage knowledge sharing. Equity alliances help only participate in a perfunctory or opportunistic manner.
because they allow partners to flexibly distribute the Leading firms can build structures to lessen part-
gains from collaboration. The firm experiencing an ners’ appropriability concerns. These structures some-
adverse shift in bargaining position can be compensated times involve Chinese walls that compartmentalize
with a greater share of the surplus. The strong partner and protect partners’ confidential information. For
can commit to a more “generous” distribution of the instance, Toyota deliberately took steps to separate its
surplus (see also Panico (2011) for a related result). operations management consulting division from its
Part (iii), however, shows that, although equity purchasing division so that suppliers would share
alliances facilitate knowledge sharing, they cannot information with Toyota without worrying that this
completely solve the problem. The reason is that shifts might constrain their future pricing flexibility.
in bargaining positions can be large relative to the Our theory, by emphasizing the broader theme of
joint surplus. Partners always retain the option to commitment to treat partners fairly (and share rents),
terminate the alliance at the negotiation stage. Thus, F also highlights the importance of the credibility of
cannot get less than πF , and A cannot get less than πA such commitments. In the case of IAL, credibility was
at the negotiation stage. But, if πF is very large, then enhanced by the fact that IAL was separated from
there may be very little surplus V − πF − πA to com- Intel’s other businesses and that this separation was
pensate A for sharing knowledge. As a result, knowl- observable from the outside. This separation, by miti-
edge sharing may not be implementable even if it is gating competitive threats from Intel’s other units,
socially efficient. increased complementors’ incentives to collaborate
with IAL. This approach also provides a different
6. Conclusion perspective on learning races. Managers should not
Firms enter into alliances for a variety of reasons: to always try to win the race but should instead some-
facilitate collusion and increase market power (Grossman times constrain their ability to appropriate value to
and Shapiro 1986), to share risks and take advantage encourage collaboration and knowledge sharing.
of new opportunities (Kogut 1991), to pool resources Alliance portfolio design can also be used to im-
with other firms (Hennart 1988) and to acquire new plicitly commit to low levels of appropriation of
skills and capabilities (Hamel 1991, Mowery et al. partners’ knowledge by creating alliance portfolios
1996, Khanna et al. 1998). with desirable characteristics. We show that the simple
In this paper, we focus on alliances in which an intuition that alliance portfolios should be constructed
important objective is the acquisition of new skills to minimize overlaps in technology among partners is
and capabilities (“learning alliances”) but in which potentially misleading because there are types of part-
contracts are incomplete and firms cannot commit to ner substitutability that may encourage rather than
exploit the newly created knowledge jointly. Firms hinder knowledge sharing. Our theory also suggests
share knowledge to create value (e.g., new products). that managers should select partners that are neither too
However, knowledge sharing also creates appropri- similar nor too dissimilar to each other because partner
ability hazards. For instance, a firm may steal a substitutability has opposing effects on value creation
partner’s trade secrets, or asymmetric learning may and value capture by the focal firm.
occur. In the latter case, the faster learner may, over Alliances are, by definition, not zero-sum games.
time, be able to reduce its dependency on the partner Instead, they have the potential to create value. Value
Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS 15

capture strategies must be balanced against the need welfare is higher in the one-sided asymmetric information scenario
to ensure value creation. Our focus on the need to than in the benchmark scenario.
13
induce knowledge sharing by the weaker partner is, In the alliance portfolio case, identifying whose knowledge-sharing
constraint is binding is less straightforward.
at the most abstract level, an attempt put the emphasis 14
Irreversibility and observability are key features of credible com-
back on value creation.
mitments (Schelling 1960, Morgan and Várdy 2013). For a commit-
ment to an action to be credible, the commitment must be irreversible,
Acknowledgments or at least, reneging must be sufficiently costly. In our model, irre-
The authors thank Catherine de Fontenay, Nick Vikander, versibility is captured by a sequential choice of actions. The as-
numerous seminar participants and especially the editors sumption that a commitment to an action is observable may not
and the reviewers for very constructive feedback and criti- always be warranted in practice. We illustrate how a commitment
cism. All remaining errors are, of course, the authors’. to protect a partner’s confidential information can be made visible
with reference to Intel’s well-publicized distinction between “job 1”
and “job 2.”
Endnotes 15
1 Allowing larger investments in absorptive capacity to also increase
Our focus on knowledge sharing also yields the counterintuitive the joint value V does not pose any difficulties, but neither does it
result that uncertainty about outcomes need not always hinder provide any insights.
knowledge sharing. Instead, greater uncertainty (in the form of mean 16
preserving spreads) or even asymmetric information may actually The model can also be extended to examine the issue of the opacity
facilitate knowledge sharing. of a firm’s operations. The learning race literature argues that firms
2 should minimize the transparency of their own operations, thus reducing
Such balanced division may result from an equitable allocation of their partners’ absorptive capacity. In terms of our model, F, instead
property rights as in Panico’s (2017) case of “symmetric control”
of increasing its own absorptive capacity (leading to a higher πF ),
or from negotiations between a monopsonistic customer and the
could reduce the transparency of its own operations (leading to a
owner of an innovation as in Aghion and Tirole’s (1994) case of
lower πA ). Arguments analogous to the ones in Proposition 4 show
“RU-ownership.”
that, in some cases, a leading firm F may find it beneficial to increase
3
Most of the property-rights literature assumes that asset ownership, the transparency of its own operations. An example would be Toyota
but not investment, affects partners’ outside options. opening its operations to General Motors to teach its partner (and
4
This paper is also related to the biform approach used in the lit- competitor) valuable lean manufacturing practices (Inkpen 2005).
erature on value creation and capture (e.g., Brandenburger and Stuart 17
On the other hand, as Lavie (2007) notes, greater overlap among
2007, Ryall and Sorenson 2007). In our model, commitments to and alliance partners increases the relative bargaining power of the focal
investments in knowledge sharing occur sequentially in the first firm. Cui (2013) argues that similarity or redundancy of resources
noncooperative stage, and payoffs result from negotiations in the among alliance partners may benefit the focal firm by securing access
second cooperative stage. to resources in uncertain environments.
5
Panico (2011) develops a model in which a focal firm selects prior to 18
By contrast, if the alliance between F and B is prematurely ter-
investment both the fraction of property rights and the fraction of minated, then B is not willing to help F with the rival project.
control rights assigned to each partner. Property rights affect the 19
Whether the alliance between F and B is prematurely terminated is
partners’ outside options, and control rights affect how the surplus
irrelevant for determining v(A). In both cases, we assume v(A)  0.
from collaboration is divided between the partners. Panico (2011)
finds that, in equilibrium, ownership and control rights are often
20
The case sRB  πF captures a situation in which either A’s or B’s
substitutes. We obtain a similar result in Section 5, in which we con- knowledge is completely redundant in creating rival benefits. For
sider equity alliances. However, in addition to contractual solutions, instance, if a large mass of AI scientists was necessary to create
our paper also considers organizational commitments (restrictions to substantial rival benefits, the partners are complements in rival
one’s own absorptive capacity and alliance portfolio design). benefits.
6
Our analysis focuses only on a few specific contracts; other contracts
21
Indeed, sImp ∈ [0, V − πF ] implies that coalitional values v satisfy
may perform better. See Segal and Whinston (2013) for a discussion of superadditivity: v(S ∪ T) ≥ v(S) + v(T) for any two disjoint set of
when/how contracts can ameliorate the holdup problem. firms S and T.
7
Veugelers and Kesteloot (1994, 1996) examine how knowledge
22
To compute φF , one can proceed as follows. There are six possible
spillovers between partners affect the stability of alliances. Although ways in which firms F, A, and B can be ordered: FAB, FBA, AFB,
these models capture the idea that knowledge spillovers undermine BFA, ABF, and BAF. The marginal contribution of F when F is the
alliance stability, they do not investigate the role of commitments in first firm to join the grand coalition (orderings FAB and FBA) is
facilitating knowledge sharing ex ante. v(F) − v(Ø)  2πF − sRB . The marginal contribution of F when F is the
8 second firm to join the grand coalition (orderings AFB and BFA) is
The assumption that knowledge sharing costs 12 I are the same for F
v(AF) − v(A)  v(BF) − v(B)  V + πF + sImp . The marginal contribu-
and A is just for simplicity. This assumption could easily be relaxed
tion of F when F is the third firm to join the grand coalition (orderings
without changing any of the qualitative results of the paper.
ABF and BAF) is v(ABF) − v(AB)  v(BAF) − v(BA)  2V + b. Because
9
It could also be that πA < 0 if the development of these alternative all orderings are equiprobable, we obtain Equation (9). φA and φB are
technologies weakens Nypro’s competitive position in the marketplace. computed similarly.
10
This is not sufficient for knowledge sharing because, if F believes 23
If α  12, then F’s equilibrium payoff is independent of Ψ and equal
that A will not invest, then F will not invest even if V − I ≥ |πF − πA |. to V − I + v + 13 b.
Here we ignore this coordination problem. 24
An equivalent way of describing the partners’ payoffs is to define a
11
We assumed I > 0 to rule out some knife-edge equilibria and contract φ such that ΠEF  (1 − φ)V − 12 I and ΠEA  φV − 12 I subject to
simplify the exposition. The analysis would remain essentially un- the constraints that (1 − φ)V ≥ πF and φV ≥ πA . These constraints
changed if we had assumed I ≥ 0 instead. define upper and lower bounds to A’s equity share φ, which ensure
12
In the proof of Proposition 3 in Online Appendix A, we provide a that partners have no incentives to prematurely terminate the alliance
precise characterization of the set of parameter values for which after knowledge sharing. The case in which φ reaches its lower bound
 Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
16 Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS

corresponds to the case in which λ  0 so that all the surplus goes to Herbst P, Walz U (2017) The design of vertical R&D collaborations.
firm F. The case in which φ reaches its upper bound corresponds to Economica 84(333):54–77.
the case in which all the surplus goes to firm A (λ  1). Iansiti M, Levien R (2004) The Keystone Advantage: What the New
Dynamics of Business Ecosystems Mean for Strategy, Innovation, and
References Sustainability (Harvard Business Press, Boston).
Inkpen AC (2000) A note on the dynamics of learning alliances:
Aghion P, Tirole J (1994) The management of innovation. Quart. J. Econom.
Competition, cooperation, and relative scope. Strategic Man-
109(4):1185–1209.
agement J. 21(7):775–779.
Argote L, McEvily B, Reagans R (2003) Managing knowledge in
Inkpen AC (2002) Learning, knowledge management, and strategic
organizations: An integrative framework and review of emerg-
alliances: So many studies, so many unanswered questions.
ing themes. Management Sci. 49(4):571–582.
Contractor FJ, Lorange P, eds. Cooperative Strategies and Alliances
Arora A, Merges RP (2004) Specialized supply firms, property rights
(Elsevier Science, Oxford, United Kingdom), 267–289.
and firm boundaries. Indust. Corporate Change 13(3):451–475.
Inkpen AC (2005) Learning through alliances: General Motors and
Bae J, Gargiulo M (2004) Partner substitutability, alliance network
NUMMI. California Management Rev. 47(4):114–136.
structure, and firm profitability in the telecommunications in-
Inkpen AC, Beamish PW (1997) Knowledge, bargaining power, and
dustry. Acad. Management J. 47(6):843–859.
the instability of international joint ventures. Acad. Management
Baum JA, Calabrese T, Silverman BS (2000) Don’t go it alone: Alliance
Rev. 22(1):177–202.
network composition and startups’ performance in Canadian
biotechnology. Strategic Management J. 21(3):267–294. Katila R, Rosenberger JD, Eisenhardt KM (2008) Swimming with
sharks: Technology ventures, defense mechanisms and corpo-
Brandenburger A, Stuart H (2007) Biform games. Management Sci.
rate relationships. Admin. Sci. Quart. 53(2):295–332.
53(4):537–549.
Christensen CM (1994) Molding the Impossible: The NYPRO/Vistakon Khanna T, Gulati R, Nohria N (1998) The dynamics of learning al-
Disposable Contact Lens Project (Harvard Business Press, Boston). liances: Competition, cooperation, and relative scope. Strategic
Management J. 19(3):193–210.
Cohen WM, Nelson RR, Walsh JP (2000) Protecting their intellectual
assets: Appropriability conditions and why US manufacturing Kogut B (1991) Joint ventures and the option to expand and acquire.
firms patent (or not). NBER Working Paper No. 7552, National Management Sci. 37(1):19–33.
Bureau of Economic Research, Cambridge, MA. Lavie D (2007) Alliance portfolios and firm performance: A study of
Cui AS (2013) Portfolio dynamics and alliance termination: The value creation and appropriation in the US software industry.
contingent role of resource dissimilarity. J. Marketing 77(3):15–32. Strategic Management J. 28(12):1187–1212.
Cui AS, O’Connor G (2012) Alliance portfolio resource diversity and McEvily SK, Das S, McCabe K (2000) Avoiding competence substi-
firm innovation. J. Marketing 76(4):24–43. tution through knowledge sharing. Acad. Management Rev. 25(2):
de Fontenay CC, Gans JS (2005) Vertical integration in the presence of 294–311.
upstream competition. RAND J. Econom. 36(3):544–572. Mitchell W, Singh K (1996) Survival of businesses using collaborative
Dussauge P, Garrette B, Mitchell W (2000) Learning from competing relationships to commercialize complex goods. Strategic Man-
partners: Outcomes and durations of scale and link alliances in agement J. 17(3):169–195.
Europe, North America and Asia. Strategic Management J. 21(2): Morgan J, Várdy F (2013) The fragility of commitment. Management
99–126. Sci. 59(6):1344–1353.
Dyer JH, Nobeoka K (2000) Creating and managing a high- Mowery DC, Oxley JE, Silverman BS (1996) Strategic alliances and
performance knowledge-sharing network: The Toyota case. interfirm knowledge transfer. Strategic Management J. 17(S2):77–91.
Strategic Management J. 21(3):345–367. Mowery DC, Oxley JE, Silverman BS (2002) The two faces of partner-
Gawer A, Cusumano MA (2002) Platform Leadership: How Intel, specific absorptive capacity: Learning and cospecialization in
Microsoft, and Cisco Drive Industry Innovation (Harvard Business strategic alliances. Contractor FJ, Lorange P, eds. Cooperative
School Press, Boston). Strategies and Alliances (Elsevier Science, Oxford, UK), 291–319.
Goerzen A, Beamish PW (2005) The effect of alliance network di- Nicita A, Sepe SM (2012) Incomplete contracts and competition:
versity on multinational enterprise performance. Strategic Man- Another look at Fisher Body/General Motors? Eur. J. Law
agement J. 26(4):333–354. Econom. 34(3):495–514.
Grindley P, Mowery DC, Silverman B (1994) SEMATECH and col- Oxley JE (1997) Appropriability hazards and governance in strategic
laborative research: Lessons in the design of high-technology alliances: A transaction cost approach. J. Law Econom. Organ.
consortia. J. Policy Anal. Management 13(4):723–758. 13(2):387–409.
Grossman GM, Shapiro C. 1986. Research joint ventures: An antitrust Oxley J, Wada T (2009) Alliance structure and the scope of knowledge
analysis. J. Law, Econom., and Organ. 2(2):315–337. transfer: Evidence from U.S.-Japan agreements. Management Sci.
Grossman SJ, Hart OD (1986) The costs and benefits of ownership: 55(4):635–649.
A theory of vertical and lateral integration. J. Political Econom. Panico C (2011) On the contractual governance of research collabo-
94(4):691–719. rations: Allocating control and intellectual property rights in the
Hamel G (1991) Competition for competence and inter-partner shadow of potential termination. Res. Policy 40(10):1403–1411.
learning within international strategic alliances. Strategic Man- Panico C (2017) Strategic interaction in alliances. Strategic Management
agement J. 12(S1):83–103. J. 38(8):1646–1667.
Hansen MT (2002) Knowledge networks: Explaining effective Pisano GP (1989) Using equity participation to support exchange:
knowledge sharing in multiunit companies. Organ. Sci. 13(3): Evidence from the biotechnology industry. J. Law Econom. Organ.
232–248. 5(1):109–126.
Hart O, Moore J (1990) Property rights and the nature of the firm. Reagans R, McEvily B (2003) Network structure and knowledge
J. Political Econom. 98(6):1119–1158. transfer: The effects of cohesion and range. Admin. Sci. Quart.
Hennart J-F (1988) A transaction cost theory of equity joint ventures. 48(2):240–267.
Strategic Management J. 9(4):361–374. Ryall MD, Sorenson O (2007) Brokers and competitive advantage.
Hennart J-F, Roehl T, Zietlow D (1999) ‘Trojan horse’ or ‘workhorse’? Management Sci. 53(4):566–583.
The evolution of U.S.-Japanese joint ventures in the United Schelling T (1960) The Strategy of Conflict (Harvard University Press,
States. Strategic Management J. 20(1):15–29. Cambridge, MA).
Arora, Belenzon, and Patacconi: Knowledge Sharing in Alliances
Management Science, Articles in Advance, pp. 1–17, © 2020 INFORMS 17

Segal I, Whinston MD (2013) Property rights. Gibbons R, Roberts J, Wassmer U, Dussauge P (2011) Network resource stocks and flows:
eds. Handbook of Organizational Economics (Princeton University How do alliance portfolios affect the value of new alliance
Press, Princeton, NJ), 100–158. formations? Strategic Management J. 33(7):871–883.
Steinhilber S (2008) Strategic Alliances: Three Ways to Make Them Work Williamson OE (1975) Markets and Hierarchies (The Free Press,
(Harvard Business Press, Boston). New York).
Swaminathan V, Moorman C (2009) Marketing alliances, firm net- Williamson OE (1983) Credible commitments: Using hostages to
works, and firm value creation. J. Marketing 73(5):52–69. support exchange. Amer. Econom. Rev. 73(4):519–540.
Vasudeva G, Anand J (2011) Unpacking absorptive capacity: A study Yan A, Gray B (1994) Bargaining power, management con-
of knowledge utilization from alliance portfolios. Acad. Man- trol, and performance in United States–China joint ven-
agement J. 54(3):611–623. tures: A comparative case study. Acad. Management J. 37(6):
Veugelers R, Kesteloot K (1994) On the design of stable joint ventures. 1478–1517.
Eur. Econom. Rev. 38(9):1799–1815. Zeng M, Hennart JMA (2002) From learning races to cooperative
Veugelers R, Kesteloot K (1996) Bargained shares in joint ventures specialization: Toward a new framework for alliance man-
among asymmetric partners: Is the Matthew effect catalyzing? agement. Contractor FJ, Lorange P, eds. Cooperative Strategies
J. Econom. 64(1):23–51. and Alliances (Elsevier Science, Oxford, UK), 189–210.